The need to better utilize medical information is constantly increasing. The main cause of this trend is the economic pressure to reduce the cost of medical care while preserving and enhancing its quality. One proposal for solving this dilemma is to develop medical information systems that assist physicians and clinical researchers in the analysis and use of the available information. A central problem that has retarded the development of useful medical information systems is the lack of suitable methods for representing clinical data. The central goal of this project is to evaluate Event Definitions (EDs) as a potential representation model for clinical data. The ED model is based on a conceptual view that a patient's medical record is a sequence of clinical events. The representation capability of the ED model will be tested in two experiments, both in the domain of chest radiology. The first experiment evaluates if the model can be used for intervocabulary translation. We want to know if the semantic nature of the EDs will provide the underlying structure necessary to map semantically related concepts. The second experiment evaluates the representation of free text. Chest x-ray reports will be abstracted in a set of clinically relevant questions with their appropriate answers. The same reports will be represented using the model. We plan to measure the information loss as the difference in the number of questions that can be answered from the ED representation versus from the narrative form.
